View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.


A review from the
Federal Reserve Bank
of Chicago

A s u p p ly -s id e e x p la n a tio n o f
E u ro p e a n u n e m p lo y m e n t
P e r fo r m a n c e a n d a c c e s s to g o v e r n m e n t
g u a ra n te e s : T h e c a s e o f s m a ll b u s in e s s
in v e s tm e n t c o m p a n ie s

A supply-side explanation of
European unem ploym ent...................................................................................... 2
Lars Ljungqvist and Thom as J. Sargent

This article offers a supply-side explanation of striking
patterns in unemployment rates and duration of unemploy­
ment in European countries, compared with other member
countries of the OECD (Organization for Economic
Cooperation and Development). The rise in long-term
unemployment in Europe is attributed to the adverse
incentive effects of generous welfare programs in times of
economic turbulence.

Perform ance and access to governm ent
guarantees: The case of small business
investm ent com panies...........................................................................................16
Elijah B rew er III, Hesna G enay,
W illiam E. Jackson III, and
Paula R. W o rth in g to n

This article analyzes the performance of small business
investment companies (SBICs) that are chartered and
regulated by the Small Business Administration (SBA).
Our principal finding is that poor performance over
the 1986-91 period is associated with high usage of
funds from the SBA.


) i\( )

!Y11( j PLRSPK( j I IV KS


hael H. Moskow
anior Vice President and D ire c to r o f Research


V’lliam C. Hunter


aarch Department
mcial Studies

; ' jglas Evanoff, Assistant Vice President
>!. ;roeconomic Policy

Charles Evans, Assistant Vice President

microeconomic Policy

Janiel Sullivan, Assistant Vice President
S -gional Programs

William A. Testa, Assistant Vice President

Anne Weaver, Manager

Helen O’D. Koshy

Rita Molloy, Kathryn Moran, Yvonne Peeples,
Roger Thryselius, Nancy Wellman

Septem ber/O ctober 1996, V olum e XX, Issue 5

the Research Department of the Federal Reserve
Bank of Chicago. The views expressed are the
authors’ and do not necessarily reflect the views of
the management of the Federal Reserve Bank.
Single-copy subscriptions are available free of
charge. Please send requests for single- and
multiple-copy subscriptions, back issues, and
address changes to the Public Information Center,
Federal Reserve Bank of Chicago, P.O. Box 834,
Chicago, Illinois 60690-0834, or telephone
(312) 322-5111.
Articles may be reprinted provided the source is
credited and the Public Information Center is sent a
copy of the published material.
ISSN 0164-0682

A supply-side explanation of
European unemployment

Lars L ju n g q vist and T h o m as J. S arg e n t

In this article, we offer a
supply-side explanation of
two striking patterns in Euro­
pean unemployment as com­
pared with that of other mem­
ber countries of the OECD (Organization for
Economic Cooperation and Development).1
(See figure 1 and table 1.) The first pattern is
that of average unemployment rates. These
were similar for European and non-European
OECD countries during the 1960s and 1970s,
but in the 1980s and 1990s average unemploy­
ment in Europe has persistently exceeded the
average in the OECD by about 2 percentage
points. Second, since the 1980s, the average
duration of unemployment in Europe has great­
ly exceeded that in the rest of the OECD. We
attribute these patterns to the incentive effects
on labor supply of unemployment compensa­
tion arrangements, which are far more gener­
ous in Europe than in the rest of the OECD.3
However, this view is challenged by the obser­
vation that unemployment compensation arrange­
ments have been more generous in Europe
throughout the post-World War II period,
during the first part of which European unem­
ployment was not higher than that for the rest
of the OECD. We attribute the rise in unem­
ployment in Europe after 1980 to a change in
the environment that required increased adapt­
ability of those workers forced to change jobs.
We show that during tranquil times, with
less need for adaptability, unemployment rates
were the same with a generous unemployment
compensation system as they would have been
without such a system. However, in turbulent
times, when greater adaptability is required, a
generous unemployment compensation system


could propel the economy into a state of persis
tently high unemployment.3
The economic environment is generally
perceived to have become more turbulent in
the last two decades. The OECD (1994) sums
it up as follows:
In the stable post-World War II eco­
nomic environment, standards of living in
most OECD countries grew rapidly, nar­
rowing the gap with the area’s highest per
capita income country, the United States.
The OECD area’s terms of trade evolved
favorably; trade and payments systems
were progressively liberalized, without
major problems; GDP and international
trade grew strongly.
In the 1970s, the economic environ­
ment became turbulent. The two oil
price rises, in 1973/74 and 1979/80,
imparted major terms-of-trade shocks,
each of the order of 2 percent of OECDarea GDP, and each sending large rela­
tive price changes through all OECD
economies. Exchange rates became vola­
tile after the breakdown of the Bretton
Woods system of fixed exchange rates.
Then there came, mainly in the 1980s,
waves of financial-market liberalization
and product market deregulation which
greatly enhanced the potential efficiency
of OECD economies, and also accelerated
Lars Ljungqvist is a senior econom ist at the Federal
Reserve Bank of Chicago and Thom as J. Sargent
is professor of econom ics at the U niversity of
Chicago, Senior Fellow at the Hoover Institutio n,
S tanford U n iversity, and a con sulta nt to the
Federal Reserve Bank of Chicago. The authors
thank Cristina deNardi fo r assistance w ith the
com putations.


show that greater earnings instabil­
ity for individual workers accom­
Unemployment rate in OECD as a percent
panied the widening earnings distri­
of the labor force
bution in the U.S. labor market,
especially in the 1980s. In fact,
half of the increased variance in
earnings for white males can be
attributed to transitory shocks that
die out within three years.
Thus, we attribute the diverse
unemployment rates observed in
Europe and the rest of the OECD
to the supply side of the labor
market and a changing economic
environment. We focus on how
mechanisms intended to provide
social insurance also encourage
people not to work. A threat of
adverse incentives haunts the deliv­
Sources: Data for 1961-77 are from OECD, L a b o r Force S ta tistics (1984)
ery of social insurance and this
and data for 1978-94 are from OECD, E m p lo y m e n t O u tlo o k (1995).
threat becomes larger in times of
economic turbulence. Social
insurance works best when exposure to an
the pace of change. All these develop­
event cannot be affected by the insured person
ments challenged the capacity of econo­
(for example, acts of nature). Our starting point
mies and societies to adapt. At the same
is that unemployment is only partly an act of
time, the need to adapt was heightened
nature, beyond the control of the worker. A
by pervasive technological change, espe­
worker makes efforts to leave a state of unem­
cially as the new information technolo­
ployment, and these efforts are influenced by
gies appeared, and by the trend towards
arrangements for compensating the worker for
being unemployed.
Gottschalk and Moffitt (1994) and Moffitt
We use a search model that views the job
and Gottschalk (1995) provide empirical evi­
market as an information processing machine.
dence of increased economic turbulence. They


Standardized unemployment rates and long-term unemployment
of 12 months or more in OECD
L ong-term u n e m p lo y m e n t
as percent of u n e m p lo y m e n t

A verag e u n e m p lo y m e n t rate
1 974-79











G erm any
U nited K ingdom








OECD Europe
Total OECD








“Data for 1980.
bAverage of data for 1979 and 1980.
Sources: The data are from OECD, Em ploym ent Outlook (1991), table 2.7, except for long-term
unemployment in 1970 which are from OECD, Em ploym ent Outlook (1983), table 24.




The market tracks and sorts infor­
mation used to match workers and
jobs. Workers and jobs have diverse
characteristics, and it is costly but
valuable to find good matches.
Market economies decentralize
job matching, leaving firms to
post vacancies and make offers
and workers to accept or reject job
offers. From both social and pri­
vate viewpoints, the state of unem­
ployment—waiting for something
better—is partly an investment in
the future.


Wage offer distribution

A search model of

Our work extends John McCall’s
(1970) basic search model to cap­
ture the effects we think differenti­
ate Europe from the rest of the OECD. Our
model is more complicated than McCall’s and
must be analyzed with a computer. However,
many of the basic insights can be conveyed by
describing first a graphical version of McCall’s
model of a reservation wage, then a graphical
version of a model of search intensity.
McCall’s basic model confronts an unem­
ployed worker with choices about employment
status and focuses on the incentives that the
market and the state present to the worker.
The model provides a framework for under­
standing how different policies affect incen­
tives and outcomes.
Balancing benefits and costs of search
Each period, an unemployed worker draws
a wage offer from a probability distribution of
offers and decides whether to accept or reject
it. Figure 2 shows a distribution of wage offers.
We let F{w) denote the probability that a ran­
domly drawn offer is less than or equal to a
given wage, vv, and the wage offer exceeds w
with probability 1-F(w). In the simplest mod­
el, an offer provides the worker with the oppor­
tunity to work indefinitely at the drawn wage.
The model also assumes that an unemployed
worker receives unemployment compensation
in a fixed amount per period for as long as he
or she is unemployed. The worker’s optimal
policy is to set a reservation wage, at a level at
which the worker is indifferent about accepting
or rejecting an offer, then to reject offers fall­
ing short of the reservation wage, and to accept
the first offer exceeding it. The model equates


being unemployed with waiting for an accept­
able offer. The worker compares the benefits
of accepting an offer with the benefits of refus­
ing it, remaining unemployed, and searching
again next period. The benefits of refusing
comprise any unemployment compensation
the worker receives this period, plus the option
value of searching again next period. The
option value covers the possibility that the
worker might eventually draw a better wage
offer in the next period or a subsequent period.
Given a particular distribution of wage
offers, we can compute and plot the present
value of all benefits associated with a policy of
setting a reservation wage of vv. To compute
present values, we let r denote the one-period
interest rate. Any benefits in the next period
can then be expressed in today’s value (present
value) when multiplying by the one-period
discount factor, B = -----. Let vF(w) be the

total benefits of rejecting a job offer today,
while setting a reservation wage of vv for ac­
cepting a job in the future. By accounting for
the various possibilities and weighting the
associated payoffs by the probabilities of oc­
currence, we can compute H'(vv) as follows:
1) H'(vv) = y + (1 - F'(vv)) ~^

+ F(w)P*F(w),
where y is the level of unemployment compen­
sation per period, and E- (w) is the expected, or


average, value of all wages exceeding a reser­
vation wage of vv.4
The right side of equation 1 expresses the
total benefits, ^(vv), as the sum of three terms:
1) y, the unemployment compensation to be
received this period; 2) the expected present
value from next period onwards of receiving
a wage exceeding the reservation wage,
£_(w)/r, weighted by the probability (1 - F(w)) of
receiving an offer next period exceeding vv;
and 3) the value of restarting the search process
next period, discounted one period by (3, and
weighted by the probability F(w), of not draw­
ing an acceptable offer next period. Equation 1
can be rearranged to become
y + ( l -F(vv))

'P (w ) =


Each curve shows how the benefits of the
search vary with the reservation wage. For low
values of the reservation wage, the benefits
increase as the reservation wage increases, but
they eventually fall for higher values of the
reservation wage. In other words, the unem­
ployed worker is better off choosing a reserva­
tion wage that is neither too low nor too high.
A too low reservation wage is not optimal,
since the worker would, on average, do better
by searching more for a somewhat higher
wage. On the other hand, a too high reserva­
tion wage does not maximize benefits, since
the worker is then, on average, spending too
much time pursuing the rare opportunity of
getting a very high wage. By setting the deriv­
ative of H'(vv) to zero, we find that the optimal
value of the reservation wage must satisfy

1 - (3F(vv)

The optimal choice of reservation wage, vv,
is the one that maximizes total benefits, H^vv).
For the same wage distribution shown in
figure 2, figure 3 plots the right-hand side of
equation 2 for unemployment compensation, y,
equal to zero and greater than zero. Since
unemployment compensation enters positively
in equation 2, the curve with some unemploy­
ment compensation is higher than the curve
without any unemployment compensation.

3) H'(vv) =


The term vv(1 +r) is the benefit of accept­
ing a wage vv immediately (that is, the present
value of receiving a wage vv today and for all
future periods). Thus, equation 3 says that the
worker optimally sets the reservation wage to
equate the total benefits of further search to the
total benefits of immediately accepting a wage
offer equal to the reservation wage. In other
words, the worker is indifferent
between continuing the search
and accepting a wage offer that is
Expected present value of payoffs for different
exactly equal to the optimal reser­
reservation wages with and without
vation wage. Figure 3 confirms
unemployment compensation
equation 3 graphically. In figure
present value of payoffs
3, for each level of unemploy­
ment compensation, the curve
showing ^ |

reservation wage
Notes: Stars indicate the optim al reservation w age for each
unem ploym ent com pensation regim e. The dashed line shows the
present value of different constant w age streams beginning today.



^ intersects total

benefits vF(vv) at the highest value
of ^(vv) (as indicated in the figure
by a star). Figure 3 shows how
an increase in unemployment
compensation increases the res­
ervation wage, because it shifts
upward the curve of benefits of
further search.
The reservation wage deter­
mines the probability of rejecting
a job offer by summing probabili­
ties attached to wage offers below
the reservation wage (see figure 2).


The rejection probability F(w) determines the
mean duration of unemployment via the formula
Duration = ----- ?------ .
1 - F(w)
Increases in F(w) increase the mean duration
of unemployment. We study how particular
policy and environmental features impinge
on the reservation wage and the duration of
Variable search intensity
The basic search model assumes that one
offer arrives per period, irrespective of the
intensity of the worker’s job search. We modi­
fy the model to let the worker influence the
probability of getting a job offer by selecting
the intensity of his or her search. To indicate
the main factors affecting search intensity, we
temporarily assume that the wage distribution
is concentrated at a point, denoted w, so that all
jobs pay the same wage w. With this assump­
tion, the only uncertainty becomes whether a
job offer arrives in the period.
We suppose that a worker chooses a prob­
ability, n, that an offer will arrive in a given
period, by incurring a utility cost, c(7t), per
period. We assume that the cost function c(7t)
has positive and increasing marginal costs:
c(7t) = 0, c'(n) > 0,
> 0. If an unem­
ployed worker decides to search this period, he
or she receives unemployment compensation
and incurs search costs of c(7t) this period. The
worker then receives an offer with probability
7t at the beginning of next period or no offer
with probability (1-71). We let Q(7t) denote the
expected present value of searching with inten­
sity 7t. We can compute
4) Q(7C) = —C(7C) + Y+ 71 — + (1—7t) (3Q(7t).
The right side of equation 4 expresses the
benefits associated with search intensity n as
the sum of four terms: 1) -c(n), the negative
value of the search cost in the current period;
2) y, the unemployment compensation to be
received in this period; 3) the present value
from next period onwards of receiving a wage

probability (1- 7t) of not drawing an offer next
period. Equation 4 can be rearranged to become
-c(7t) + y + 7t —

The optimal choice of probability, 7t, is the one
that maximizes total benefits, Q(n).
Figure 4 displays the three components of
the right side of equation 5 as functions of the
probability of getting a wage offer, while figure
5 displays their sum for two different levels of
unemployment compensation, y, equal to zero
and greater than zero. (These graphs assume the
particular cost function c(n) = 50k 4.) As shown
in figure 4, for a given level of unemployment
compensation, the expected present value of
received unemployment compensation decreas­
es as the probability of an offer increases.
Moreover, the higher the level of unemploy­
ment compensation, the higher this curve in
figure 4. It follows that the higher the level of
unemployment compensation, the lower the
probability of getting a wage offer (correspond­
ing to a lower search intensity) that maxi­
mizes the total benefits. Figure 5 shows
how the optimal setting of the probability
declines as unemployment compensation
increases. (We mark the optimal probability
for each level of unemployment compensa­
tion with a star.)
In this setting, the average duration of
unemployment is just
. By causing the
probability (tu) to decrease, increases in unem­
ployment compensation cause the mean dura­
tion of unemployment to rise. Similar forces
operate in the more general model when the
distribution of offers is nontrivial. The main
difference is that the value of a wage offer in
the above equations must be replaced with a
value that depends on the worker’s reservation
wage, which is also influenced by the level of
unemployment compensation. This is the case
we are interested in.
Extensions of the basic search model

receiving an offer next period; and 4) the value
of restarting the search process next period,
discounted one period by [5, and weighted by the

To construct our theory of European un­
employment, we add three features to the basic
search model outlined above—job termination,
human capital/skills, and eamings-dependent
unemployment compensation.
Job termination—We have adjusted the
option value of searching for a job to reflect



weighted by the probability, n, of

We let human capital appreciate
when the worker is employed,
Expected present values of wages, search costs,
and let it depreciate gradually
and unemployment compensation
during spells of unemployment.
(as given by the three components in equation 5)
Human capital/skill levels differ­
expected present values
entiate workers from each other;
unemployed workers with differ­
ent human capital levels set dif­
ferent reservation wages and
search intensities.
We specify a given number
of potential levels of human capi­
tal or skills, ordered from lowest
to highest. We also specify two
sets of transition probabilities,
describing the change in skills
over time. For example, we
would expect a worker’s skills
to improve during periods of
employment and, conversely, to
deteriorate during periods of
the possibility that an existing job terminates
We define a worker’s total earnings as the
against the will of the worker. Exposing the
product of a base wage, to be drawn from a
given wage offer distribution, and the worker’s
worker to a small probability of involuntary
skills. During a spell of employment, a worker
job loss each period tends to diminish the op­
tion value of a further job search, and can di­
who starts from a low level of skills can expect
minish the reservation wage.
his or her earnings to grow gradually as his or
Human capital or skills—We have made
her skills grow, even though the base wage is
earnings depend on human capital or skills.
set once-and-for-all at the beginning of the
current spell of employment.
The worker takes into account the
likely growth of earnings in for­
mulating the reservation wage
Expected total payoffs with and without
unemployment compensation
and search intensity. The worker
(as given in equation 5)
also takes into account the way
unemployment compensation
expected present value of payoffs
depends on past earnings.
Earnings-dependent unem­
ployment compensation—The
basic model has a fixed level of
unemployment compensation,
independent of the worker’s earn­
ings during previous employment.
We modify this feature by linking
unemployment compensation to
earnings attained on the previous
job. This means the option value
of the search will depend on the
worker’s current skill level, the
effect of prospective employment
status on the worker’s skills, and
Note: Stars indicate the optim al search intensity for each unem ploym ent
com pensation regim e.
the level of the worker’s previous
earnings. The effect on this option





value of unemployment compensation and the
latter’s dependence on past earnings form an
important part of our analysis.
Representing economic turbulence
Our model contains two types of parame­
ters that can be used to represent labor market
turbulence, a parameter representing firing or
job dissolution and parameters governing the
rate at which human capital depreciates while
unemployed. We will use one particular
parameter from the latter set to measure turbu­
lence, namely a parameter that sets the one­
time depreciation in skill level that an employed
worker experiences upon becoming unem­
ployed. In tranquil times, we let the worker
experience no immediate depreciation in human
capital, but in turbulent times, we expose the
worker to a one-time reduction in human capital.
This is our way of roughly capturing the dis­
parity in skills used in different jobs. In tran­
quil times, skills are more transferable than in
turbulent times, when job descriptions change
more quickly.
Consequences of additional features
The modifications of the basic model, in
our view, provide a more realistic picture of
the incentives unemployed workers face. Given
the possibility that a job may terminate, the
unemployed worker takes into account not only
current unemployment compensation, which is
linked to past earnings, but also the fact that
future unemployment compensation will be
linked to future earnings, which depend on the
worker’s base wage and human capital level.
Because the human capital level deteriorates
with the passage of time spent unemployed,
the worker will balance the benefits of waiting
for a higher base wage against the prospects
of further deterioration of human capital
while unemployed.
The balance will depend on the level of
unemployment compensation. High unemploy­
ment compensation sets the following trap.
Consider a worker who had relatively high
earnings before losing a job and, therefore,
qualifies for a high level of unemployment
compensation. This worker’s reservation base
wage and search intensity each depend on his
or her human capital level. Early in a spell of
unemployment, the worker searches intensive­
ly, and sets a reasonable reservation base wage,
because his or her earnings are the product of
that wage and the human capital level and,


even for typical wages, the associated earnings
compare favorably with unemployment com­
pensation. However, if the worker remains
unemployed for a while and experiences a
deterioration in human capital, the incentives
change adversely. The worker’s unemploy­
ment compensation remains high (tied to previ­
ous earnings), but for any given prospective
draw from the base wage distribution, the earn­
ings are lower because of diminished human
capital. Because the benefits of searching have
declined relative to the compensation for re­
maining unemployed, the worker will tend to
search less intensively and to set a higher reser­
vation base wage. This behavior, in turn, will
diminish the worker’s probability of leaving
unemployment and increase the mean duration
of unemployment.
Human capital acquisition can also repre­
sent a source of quits or voluntary separations.
A worker with low human capital may accept a
lower base wage than one who has higher hu­
man capital. Having subsequently experienced
growth in human capital, the worker may find
it optimal to quit the job and search for a high­
er base wage to capitalize on his or her higher
human capital.
Equilibrium: Many workers

The search model captures the experiences
of an individual worker as time and opportuni­
ties pass. We can use it as a building block to
model the behavior of a large number of ex ante
identical but ex post diverse workers composing
a complete labor market. To build a model of
the labor market, we reinterpret the search model’s
individual descriptive statistics—average dura­
tion of unemployment, average accepted wage,
average times between incidents of quitting or
being fired—as applying to the average at any
point in time of a large number of statistically
identical individuals.
Imagine the labor market as a set of lakes
connected by inlet and outlet streams (see
figure 6). The volume of water in each lake
represents the number of people in a particular
labor market state (for example, employed and
unemployed with different levels of human
capital), and the flows between lakes represent
rates of hiring, firing, and quitting. The system
is in equilibrium when all lake levels are con­
stant over time, which means that inflows
balance outflows for each lake. The rates of
inflow and outflow are the critical determinants


of the lake levels. The individual search model
lends itself to becoming a model of these inflow
and outflow rates. For example, we can interpret
the probability of job acceptance as determining
the rate of flow from a state of unemployment
to a state of employment.
Within such a model, government-supplied
unemployment compensation gives rise to
expenditures that must be financed. In particu­
lar, the size of the unemployment lake (or lakes)
determines the total volume of government
unemployment compensation payments. We
suppose that these are financed from income
taxes and that, in a state of equilibrium, govern­
ment expenditure rates and tax rates must be set
so that the government budget balances.
Numerical examples

We use numerical simulations to illustrate
the equilibrium forces at work in our aggregate
model of the labor market. Our results are
mainly driven by two sets of parameters—the
skill technology and the unemployment com­
pensation scheme.6
Our model includes 21 skill levels and
assumes that all new entrants to the labor mar­
ket start out with the lowest skill level. After
each two-week period of employment that is
not followed by a layoff, the worker has a one
in four chance to increase skills by one level;



otherwise, the skill level remains
unchanged. Employed workers
who have reached the highest
skill level retain those skills until
becoming unemployed. It will
take a worker who is continuously
employed, on average, about
three years and one month to
reach the highest skill level. We
assume that the stochastic depre­
ciation of skills during unemploy­
ment is twice as fast as the accu­
mulation of skills. That is, after
each two-week period of unem­
ployment, there is a one in two
risk that the worker’s skills de­
crease by one level; otherwise,
they remain unchanged. Once the
lowest skill level is reached through
depreciation, the worker remains
at that level until becoming em­
ployed. Finally, in a period of
being laid off, it is assumed that
the worker keeps the skill level
from the last period of employment. As pointed
out above, our definition of tranquil economic
conditions implies that skill depreciation is
related only to the time spent unemployed;
there is no unusual loss of skills associated
with the layoff itself.
Figure 7 depicts a random realization of
skills for a new entrant into the labor market.
The vertical dotted lines separate periods of
employment and unemployment. According to
the figure, the worker’s first job lasts for almost
two years, during which he or she accumulates
considerable skills. However, the following
3.5 month spell of unemployment is associated
with skill depreciation. After finding a second
job, the worker remains there for three years
and attains the highest skill level. Following
another short spell of unemployment, the
worker finds a third job and regains the skills
lost during unemployment.
Concerning the unemployment compensa­
tion scheme, we examine the outcome for two
economies, one with unemployment insurance
and one without. The economy with unem­
ployment insurance is called the welfare state
(WS) and has a 70 percent replacement ratio,
that is, unemployment benefits cover 70 percent
of lost earnings for laid off workers.7 The econ­
omy with no unemployment insurance is called
the laissez-faire (LF) economy.


proportion of long-term unem­
ployed at any point in time. In
Random realization of a worker’s skill level
the WS economy, 14.0 percent of
under tranquil economic conditions
currently unemployed workers
skill level
have unemployment spells to date
greater than or equal to six months
(and 5.1 percent greater than or
equal to 12 months), compared
with 4.7 percent (0.3 percent) in
the LF economy. However, in
absolute numbers these long-term
unemployed workers constitute a
very small portion of the total
labor force; a 3.3 percent income
tax is sufficient to finance the
unemployment insurance scheme
in the WS economy.
To understand why the equi­
libria in these two economies are
Note: The vertical dotted lines separate periods of em ploym ent (skill
virtually the same, we make a
accum ulation) and unem ploym ent (skill depreciation).
connection between the workers’
behavior, as discussed earlier, and
the economy’s aggregate perfor­
mance. Let us track a large group of workers
Tranquil economic times
Table 2 reports on the equilibria under
who lost their jobs after having attained the
tranquil economic times for the WS economy
highest skill level. Although we can not deter­
and the LF economy. It might be surprising to
mine precisely the fate of each individual un­
see that the economies look very similar in
employed worker (since luck plays a role in
terms of unemployment levels and duration.
what wage offers an individual obtains), we
The unemployment rate is only eight-tenths of
can compute average outcomes for these work­
a percentage point higher in the WS economy.
ers as a group. Specifically, at different unem­
The average unemployment spell is 11.6 weeks
ployment durations, we can estimate the hazard
in the WS economy versus 9.4 weeks in the
rate of gaining employment, that is, the pro­
LF economy. However, the WS economy has
portion of still unemployed workers who gain
considerably more dispersion in the duration
employment in the current two-week period
of unemployment spells, as indicated by the
(see figure 8).
As shown in figure 8, the
hazard of gaining employment in
the LF economy is first increasing
and then decreasing. These dy­
Equilibria in WS economy and LFeconomy in
namics are completely driven by
tranquil economic times
changing reservation wages over
the unemployment spell. Initial­
ly, these workers with the highest
U n em p loym e nt rate (percent)
skill level have nothing more to
Average duratio n of
un em p loym e nt (weeks)
gain in terms of skills so they find
Percent of unem ployed
it optimal to search for a very
w ith spells so far > 6 m onths
good wage, that is, they choose
Percent of unem ployed
high reservation wages. As time
w ith spells so far > 12 m onths
goes by, some workers are unlucky
Tax fina ncing un em p loym e nt
in their job search and their skills
benefits (percent)
start depreciating due to unem­
Note: n.a. indicates not applicable.
ployment. It then becomes opti­
mal for them to choose a lower




find jobs paying more than their
current replacement ratio of 70
Hazard of gaining employment as a function
percent. But after this initial
of the length of the unemployment spell
period, the hazard of gaining
hazard of gaining employment
employment falls dramatically in
the WS economy. Long-term
unemployed workers in the WS
economy become disillusioned
when they experience skill depre­
ciation. In other words, the pas­
sage of time makes the prospect
of finding a job less attractive,
compared with living on unem­
ployment benefits. These workers
hold out for a very good wage
offer before giving up their gener­
ous benefits (relative to their
currently low skills). Since it is
rare to find such good wage offers,
Notes: W e assume an initial skill level equal to the highest one. The curves
they reduce their investment in
show the fraction of still unem ployed workers who gain em ploym ent in any
given tw o-w eek period after the layoff, w ith the w orkers' last earnings
search, that is, they reduce the
belonging to the most com m on income class.
intensity of their job search.
Under the assumed tranquil
these incentive problems
reservation wage, thereby increasing the
only a small impact
chance of finding an acceptable job and reduc­
The average dura­
ing the risk of further skill deterioration. This
WS economy is
accounts for the increasing segment of the
table 2. So most
hazard function in figure 8. However, a very
before becom­
small group of the unemployed workers in the
LF economy will find themselves unemployed
for more than a year (0.3 percent of all unem­
A transient economic shock
ployed, as shown in table 2). These workers
The unemployment dynamics described
will once again find it optimal to choose higher
above make the WS economy more vulnerable
reservation wages; because they have already
to economic shocks than the LF economy.
lost most of their skills, the cost of searching
This can be demonstrated by examining the
for a better wage has actually gone down.
economies’ behavior in response to a transient
Consider a similar group of laid off work­
unemployment shock. We assume that the
ers in the WS economy. Because these work­
normal layoff rate increases sharply (twentyfold)
ers receive unemployment benefits with a
in a single two-week period, and that everyone
replacement ratio of 70 percent, we have to
who becomes unemployed in this particular
make an assumption about their lost earnings.
period immediately loses 75 percent of his or
(As mentioned earlier, their choice of reserva­
her skills. After this one-period shock, both
tion wages and search intensities will depend
economies revert to their normal layoff and
on their unemployment compensation.) Figure
skill depreciation/accumulation rates. Policy
8 depicts the hazard of gaining employment in
parameters, such as taxes and the unemploy­
the WS economy, under the assumption that
ment compensation program, are kept constant
lost earnings were in the most common income
throughout the experiment. It follows that the
class. Note that the LF and WS curves in figure 8
economies will eventually return to the equilib­
are remarkably similar during the first four
ria in table 2.8
months. Despite their unemployment compen­
As shown in figure 9, the shock causes
sation, workers in the WS economy choose to
unemployment rates in both economies to jump
search for and accept jobs in similar ways to
initially by about 16 percentage points. Howev­
those in the LF economy. They are eager to
er, in the LF economy, the high unemployment





rate dies out quickly because unemployed
workers search intensively for jobs paying less
than their previous employment. In contrast,
workers in the WS economy enter a prolonged
period of unemployment because, given their
depreciated skills, they have difficulty finding
jobs that they prefer to their unemployment
compensation (which is based on past earn­
ings). Besides setting high reservation wages,


these workers also reduce search
intensities to balance the small
prospective gains with the utility
costs of search.
Panels A and B of figure 10
show how long-term unemploy­
ment gradually emerges after the
shock. At any point, the figures
decompose unemployment into
the fraction of unemployed work­
ers who have been unemployed
for at least one year, those who
have been unemployed for be­
tween six months and one year,
and those who have been unem­
ployed for less than six months.
Not surprisingly, both of the first
two measures of unemployment
fall at the time of the shock,
when there is a flood of newly
laid off workers. The two mea­
sures then rise predictably after six months
and 12 months, respectively. The problem of
long-term unemployment in the WS economy
shows up starkly in panel A of figure 10. In
contrast, the LF economy (panel B) has a much
lower incidence of long-term unemployment,
and there is hardly any persistence in the
fractions of long-term unemployed, com­
pared with the WS economy.


that a worker draws a new skill
level from one of the distribu­
tions in figure 11. The range of
each distribution starts at the
lowest possible skill level and
ends at the worker’s skill level
before the layoff. In other words,
the worker stands to lose some of
his or her skills immediately and
a few workers may even draw a
significantly lower skill level in
the left-hand tail of the distribu­
tion. During the unemployment
spell itself and at times of con­
tinuing employment, skills depre­
ciate and accumulate as before.
The new skill technology is
illustrated in figure 12 analogously
to figure 7. In fact, both figures
depict exactly the same realization
of the direction of skill movements
during unemployment spells and at
times of continuing employment.
Turbulent economic times
The only difference is that the new skill technolo­
Below, we show how the poor unemploy­
gy may give rise to additional skill losses exactly
ment performance of the WS economy in re­
at the time of layoffs. As can be seen, the extra
sponse to a transient economic shock will persist
skill loss is pretty modest at the first layoff, but
during times of ongoing economic turbulence.
the second time around the skill loss is close to
We define economic turbulence in terms of the
30 percent of the worker’s accumulated skills.
mean and variance of skill losses associated
The extra skill losses occasionally associated
with layoffs. At the time of a layoff, we assume
with job losses (figure 12) affect
the unemployed worker’s search
intensity and reservation wage, as
in the case of a transient
Random realization of a worker’s skill level under
unemployment shock above. This
turbulent economic times
means that the length of unem­
skill level
ployment spells can vary between
figures 7 and 12 because of the
different incentives confronting
these two unemployed workers.
We want to know how these
changes will affect economy-wide
average rates of unemployment
and long-term unemployment in
the WS and LF economies.
To address this question,
we compute equilibria for each
degree of economic turbulence in
Figure 11. We use the equilibria
under tranquil economic condi­
tions (discussed above) as a bench­
Note: The vertical dotted lines separate periods of em ploym ent (skill
mark case.9 As shown in figure 13,
accum ulation) and unem ploym ent (skill depreciation).
unemployment remains virtually





Unemployment rates under different
degrees of economic turbulence
unemployment rate


degree of economic turbulence


flat in the LF economy in response to increased
economic turbulence, while both the unemploy­
ment rate and the incidence of long-term unem­
ployment rise sharply in the WS economy (in the
LF economy, long-term unemployment remains
low and therefore is not visible in figure 13).
The explanation of these patterns is essentially
the same as that for the responses to a transient
economic shock. Moreover, the pressure to fi­
nance the unemployment compensation scheme
in the WS economy naturally increases with
economic turbulence. Thus, the tax rate of 3.3
percent required to finance unemployment com­
pensation under tranquil conditions (see table 2)

increases to 9.1 percent under the highest degree
of economic turbulence (indexed by .0175).

Our analysis suggests that high unemploy­
ment rates in Europe can be attributed to the
adverse incentive effects of generous welfare
programs in times of economic turbulence.
According to this view, the smooth perfor­
mance of the European welfare states up to
the 1970s was due to tranquil economic times,
while the current unemployment crisis has
been brought about by a change in the econom­
ic environment that required increased adapt­
ability of the workers forced to change jobs.
Since generous benefits based on past earnings
greatly diminish the incentives for individual
workers to accept a transition to a new job,
where skills once again have to be accumulat­
ed, our model predicts a high incidence of
long-term unemployment in the welfare states.
In fact, more than half of all those currently
unemployed in Europe have been out of a job
for more than a year.
Our analysis highlights the need to reform
European social insurance programs. This is a
real challenge, because a more turbulent eco­
nomic environment has both reduced the effec­
tiveness of existing social safety nets and in­
creased the perceived need for social insur­
ance. But the fact remains that it is more im­
portant than ever to incorporate incentives to
work in the design of social safety nets. Fail­
ure to do so threatens to produce high and
long-term unemployment and needlessly to
waste human capital.

'This article summarizes our research on European unem­
ployment, and a more detailed account can be found in
Ljungqvist and Sargent (1995).
T he notion of unemployment compensation should be
interpreted broadly in our framework. The welfare states
have various programs assisting individuals out of work.
For example, totally disabled persons in the Netherlands
in the 1980s were entitled to 70 percent (80 percent prior
to 1984) of last earned gross wage until the age of 65—
after which they moved into the state pension system. At
the end of 1990, disability benefits were paid to 14 per­
cent of the Dutch labor force and 80 percent of them were
reported to be totally disabled. (See Organization of
Economic Cooperation and Development 1992).
Tn contrast to our labor supply explanation, earlier theories
of European unemployment have focused on a shortfall in


the demand for labor due to insufficient aggregate de­
mand (Blanchard et al. 1986), trade union behavior driven
by insider-outsider conflicts (Blanchard and Summers
1986; Lindbeck and Snower 1988), hiring and firing costs
(Bentolila and Bertola 1990), and capital shortages (Malinvaud 1994). Our analysis will instead bear out the
assertion by Layard, Nickell, and Jackman (1991, p. 62)
that the “unconditional payment of benefits for an indefi­
nite period is clearly a major cause of high European
unemployment.” However, our model differs sharply
from their framework, which emphasizes hysteresis and
nominal inertia in wage and price setting.
4Formally, the conditional expectation of wages exceeding
a reservation wage, w, is given by
f_w f(w )dw


where f(w) is the probability density function for wage
offers, and F(w) = Prob(w < vv) = j0f{w)dw is the cumu­
lative density function.
5A troublesome feature of the basic search model is the
existence of the always rejected part of the wage distribu­
tion beneath the reservation wage. The presence of such
offers justifies the time the worker waits for higher ones.
But if such offers are always rejected, why do firms
continue to make them? This conceptual problem has
been circumvented by reinterpreting the wage as an
overall measure of worker-firm job match quality. Many
features influence the quality of matches between hetero­
geneous collections of workers and jobs. The idea is to
reinterpret the wage as a match parameter that aggregates
these diverse features of a job-person match. Thus, a
worker-firm pair is actually jointly drawing a match
quality each time an unemployed worker receives a job
offer. We still interpret this parameter as the wage of the
worker, but regard it as compensation for a particular
match quality. This interpretation leaves room for offers

that are rejected by one worker to be accepted by another,
because they are different matches.
6For a detailed discussion of all parameter values in our
model, see Ljungqvist and Sargent (1995).
’Workers who have quit their jobs and new entrants to the
labor market are not entitled to any benefits in our model.
Moreover, the insured unemployed workers are disquali­
fied from receiving benefits if they are discovered turning
down job offers that would have earned them at least as
much as their current unemployment compensation.
8We assume that the extra government expenditures on
unemployment compensation in the WS economy are
financed by levying lump-sum taxes, that is, nondistor­
tionary taxes.
9The tranquil economic environment has a zero variance
according to our definition of economic turbulence. Recall
that our earlier assumption was that a newly laid off worker
kept his or her skills from the last period of employment.

Bentolila, Samuel, and Giuseppe Bertola,
“Firing costs and labor demand: How bad is
Eurosclerosis?” Review of Economic Studies,
Vol. 57, No. 3, July 1990, pp. 381^102.

Ljungqvist, Lars, and Thomas J. Sargent,
“The European unemployment dilemma,” Feder­
al Reserve Bank of Chicago, working paper, No.
17, 1995.

Blanchard, Olivier J., and Lawrence H. Sum­
mers, “Hysteresis and the European unemploy­
ment problem,” in NBER Macroeconomics Annu­
al, Stanley Fischer (ed.), Cambridge, MA: MIT
Press, 1986, pp. 15-78.

Malinvaud, Edmond, Diagnosing Unemploy­
ment, Cambridge, UK: Cambridge University
Press, 1994.

Blanchard, Olivier, Rudiger Dornbusch,
Jacques Dreze, Herbert Giersch, Richard
Layard, and Mario Monti, “Employment and
growth in Europe: A two-handed approach,” in
Restoring Europe's Prosperity: Macroeconomic
Papers from the Center for European Policy
Studies, Olivier Blanchard, Rudiger Dornbusch,
and Richard Layard (eds.), Cambridge, MA: MIT
Press, 1986, pp. 95-124.
Gottschalk, Peter and Robert Moffitt, “The
growth of earnings instability in the U.S. labor
market,” Brookings Papers on Economic Activi­
ty, No. 2, 1994, pp. 217-272.
Layard, Richard, Stephen Nickell, and Rich­
ard Jackman, Unemployment: Macroeconomic
Performance and the Labor Market, Oxford, UK:
Oxford University Press, 1991.
Lindbeck, Assar, and Dennis J. Snower, The
Insider-Outsider Theory of Unemployment,
Cambridge, MA: MIT Press, 1988.



McCall, John J., “Economics of information and
job search,” Quarterly Journal of Economics, Vol.
84, No. 1, February 1970, pp. 113-126.
Moffitt, Robert A., and Peter Gottschalk,
“Trends in the autocovariance structure of earn­
ings in the U.S.: 1969-87,” Brown University
and Boston College, working paper, 1995.
Organization for Economic Cooperation and
Development, Employment Outlook, Paris:
OECD, 1995.
_________ , The OECD Jobs Study: Facts,
Analysis, Strategies, Paris: OECD, 1994.
_________ , OECD Economic Surveys—Nether­
lands, Paris: OECD, 1992.
_________ , Employment Outlook, Paris:
OECD, 1991.
_________ , Labor Force Statistics, Paris:
OECD, 1984.
_________ , Employment Outlook, Paris:
OECD, 1983.


Performance and access to government
guarantees: The case of small business
investment companies

Elijah B re w e r III, Hesna G en ay,
W illia m E. Ja ck so n III, and
P aula R. W o rth in g to n

In 1953, Congress established
the Small Business Adminis­
tration (SBA) to ensure the
provision of adequate capital
for the formation and growth
of the nation’s small businesses.1 Small busi­
ness investment companies (SBICs) are SBAchartered and -regulated financial intermediar­
ies that finance the activities of small business
through equity investments and loans. While
traditional financial intermediaries such as
commercial banks provide loans to businesses,
they do not, in general, provide equity financ­
ing. However, SBICs can simultaneously hold
the equity of and lend to a client commercial
firm. SBICs obtain their funds primarily from
two sources—privately invested capital and
long-term debentures (leverage) guaranteed by
the SBA. In this article, we analyze the perfor­
mance of 280 SBICs that were active at the
beginning of 1986. Of these 280 SBICs, over
half, or 56 percent, had failed by 1993. As of
September 1995, 189 SBICs were in liquidation,
with SBA-guaranteed debentures outstanding of
over $500 million.2 The U.S. General Account­
ing Office (GAO) estimated that only $200
million would ultimately be repaid (United
States General Accounting Office 1995).
While these absolute dollar losses are
small, the failure rates and the associated losses
per dollar of guaranteed debentures are quite
high compared with those of banks and thrifts
over the 1980-91 period.3 Because the SBA, a
government agency, provides funds directly to
SBICs and serves as a financial guarantor of

E lijah B rew er III, Hesna G enay, and Paula R.
W o rth in g to n are e co n o m ists at the Federal
Reserve Bank o f Chicago and W illia m E. Jackson
III is an assista nt p ro fe sso r at the U n iv e rs ity of
N orth C a rolina at Chapel H ill. The au thors
w ould like to thank Julian Zahalak fo r his excellent
research assistance, the S m all Business A d m in ­
istration fo r providing the data, Leonard W. Fagan,
Jr., fo r providing detailed in fo rm atio n on the SBIC
program , and Anil Kashyap and David M arshall fo r
com m ents on earlier drafts of this paper.



securities sold by SBICs to third parties, tax­
payers’ funds are at risk. As a result, policy­
makers and taxpayers have a stake in evaluat­
ing the economic performance of SBICs. Such
a study can shed light on the impact of govern­
ment subsidization and loan guarantees on the
behavior of financial intermediaries.
Furthermore, the SBIC program enlarges
the permissible activities and investments of
banking organizations beyond those typically
permitted for their commercial bank and ven­
ture capital units. Banking organizations own
and operate SBICs, as well as other venture
capital firms. While traditional bank-owned
venture capital units can only own up to 5
percent of a small firm’s equity, SBIC units of
banking organizations can own up to 50 per­
cent of a small firm’s equity. Thus, the SBIC
program gives banking organizations a way to
hold a substantial amount of commercial firms’
equity while simultaneously holding their debt.
Learning about how bank-owned SBICs oper­
ate may shed light on what could happen if the
restriction on bank ownership of shares in
commercial enterprises were relaxed.

In previous research, Brewer and Genay
(1994, 1995) studied the profitability of SBICs
and documented a negative relationship be­
tween their use of SBA leverage and returns on
equity (ROE). In this article, we extend this
work to consider the relationship between
various financial factors and SBIC failure, as
well as the relationship between those factors
and ROE, with special attention paid to the
roles played by SBA leverage and SBICs’
investment choices. We find that the relation­
ship between failure and SBA leverage is posi­
tive and that between ROE and SBA leverage
is negative. Poor short-term performance, as
measured by ROE, does not necessarily imply
losses to the taxpayers. Losses are incurred
only when an SBIC experiences sustained
losses over time and is unable to meet its obli­
gations. For this reason, we also use a long­
term measure of SBIC performance, specifical­
ly whether an SBIC fails or survives, to assess
the relationship between SBA funding and the
performance of SBICs.
Because Brewer and Genay (1994, 1995)
found evidence that bank-owned SBICs dif­
fered significantly from nonbank-owned
SBICs, we also consider whether the SBA
leverage-performance relationship differs
between bank-owned and nonbank-owned
SBICs. We find that, compared with nonbank-owned SBICs, bank-owned SBICs had
higher ROEs and lower SBA leverage use,
and their investments in small businesses
were more likely to be in equity form and to
be intended for projects requiring careful
monitoring, such as research and development
and marketing projects. We also find that the
significant negative relationship between SBA
leverage and ROE differs between the two
types of SBICs. When leverage is measured
by an SBIC’s ratio of SBA-guaranteed debt to
total assets, both bank- and nonbank-owned
SBICs exhibit a strong, negative relationship
between ROE and leverage—high leverage
use is associated with low ROE. Using an
alternative leverage measure, the ratio of
SBA-guaranteed debt to private capital, yields
similar results. But when leverage is mea­
sured by the change in SBA funding relative
to assets, the negative relationship remains
significant only for nonbank-owned SBICs.
The lack of correlation between leverage and
ROE for bank-owned SBICs holds, even


when we examine only those bank-owned
SBICs that have positive SBA leverage. This
suggests that the perceived costs and benefits
of using SBA subsidies differ across SBIC
types. Our findings for SBIC failure rates are
broadly similar to those for ROE. In particu­
lar, we find that the likelihood of an SBIC
failure increases with SBA leverage, though
our results are somewhat sensitive to the defi­
nition of failure.
Our findings that ROE decreases and the
likelihood of failure increases with SBA lever­
age are consistent with 1) the notion that risky
SBICs are more likely to make greater use of
SBA funding than other investment companies
(adverse selection)-, 2) the tendency for firms
with government liability guarantees to invest
excessively in risky assets (moral hazard)-, 3)
the prepayment effect, stemming from an SBA
restriction that limited the ability of SBICs to
refinance their SBA debt; and 4) the mismatch
effect resulting from using SBA debt to finance
equity investments. We offer some evidence
on these explanations, but we cannot defini­
tively quantify the relative importance of each.
However, our research suggests that govern­
ment subsidization of activities to fund small
businesses can have unintended consequences
if the assets financed by the subsidized interme­
diaries are riskier than they would be in the
absence of the subsidies.
The SBIC program

The SBIC program was established in
1958 and is administered by the SBA.4 The
goal of the program is to encourage the provi­
sion of long-term capital to small firms, de­
fined as firms having less than $6 million in
net worth or a two-year average net income of
less than $2 million. A company can be licensed
as an SBIC if it satisfies a minimum capital
requirement of $1 million. SBICs can be orga­
nized as corporations or partnerships and can
be owned by individuals or other firms, includ­
ing banking organizations.
Investment companies are eligible to receive
subsidized funds through the issuance of de­
bentures which are purchased directly or are
guaranteed by the SBA. These debentures are
usually of ten years duration. Each SBIC can
receive up to $3 in SBA funds for every $1 of
private capital, up to a maximum of $35 million.5
The SBA’s creditor position on debentures is fully


subordinated to all third-party creditors of the
SBIC. Furthermore, if an SBIC is organized as
a partnership, the general partner of the firm, in
general, is not liable for the debt.6 However, as
a condition of receiving funds, the SBA may
require a general partner to guarantee the repay­
ment of SBA debt. Finally, during the period
under review, SBICs could not prepay their
SBA-held or -guaranteed debt during the first
five years of issue.
SBICs provide both equity capital and
long-term loans to small firms. However,
they are subject to certain restrictions on their
investments. SBICs cannot invest in certain
sectors, such as real estate, or foreign firms,
and, in general, they cannot provide short­
term financing. If an SBIC makes an equity
investment in a small firm, it cannot acquire a
controlling interest without a plan of divesti­
ture.7 SBICs owned by banking organizations
face the same regulations on equity invest­
ments as other SBICs. The SBA also places
restrictions on the maturity and interest rate of
loans made by SBICs. The minimum maturi­
ty allowed is five years; the maximum interest
rate that can be charged to small businesses is
based on the interest rate on debentures issued
by the SBICs.8
SBICs are subject to annual examinations
by the SBA and certain reporting requirements,
such as reporting their financial condition
annually. They also are required to provide
documentation on each investment they make
in a small business. For instance, SBICs are
required to provide information certifying that
the firm meets SBA size standards and describ­
ing the financial condition of the firm.
In addition to these oversight regulations,
SBICs using SBA leverage are subject to capi­
tal requirements. The SBA determines that an
SBIC has serious financial problems if the sum
of its net realized losses plus net unrealized
losses on securities held exceeds 50 percent of
its capital. If an SBIC is capital impaired by
this test, the SBA gives the firm an opportunity
to correct its weak capital condition. If the
SBIC fails to correct the capital impairment or
defaults on its payments, the entire SBA debt
may be declared immediately payable. Under
these circumstances, or if there is another vio­
lation of the loan agreement or any agreement
with the SBA, the SBIC is liquidated or its
license is revoked.



Performance of SBICs and other
financial institutions
A. Returns on equity, 1986-91








1 1

______ 1______ 1______ 1______ 1__






B. Failure rates, 1987-93
percent of institutions active at beginning of year

Notes: Average failure rates = 11.24 percent for
SBICs; 5.50 percent for S&Ls; and 1.13 percent for
com m ercial banks. For SBICs, failure is defined as
liquidation or revocation of an SBIC's license by
the SBA or surrender of license by an SBIC.
Sources: Authors' calculations from data provided by
the U.S. Sm all Business Adm inistration (SBA), the
Office of Thrift Supervision, and in various issues of
the FDIC Q u a rte rly B an kin g P rofile.

Overview of performance and leverage

The data used in this article are for 280
SBICs active at the beginning of 1986, which
filed reports of both condition and invest­
ments.9 The reports of condition provide detailed
balance-sheet and income-statement informa­
tion of SBICs for the 1986-91 period.10 The
investment data provide the name, SIC code,
total assets, number of employees, and location
of the firms being financed; the dollar amount
and type of financing provided (loans, equity,
or debt with equity features); whether there
was a put option on the equity financing that
requires the small firm to repurchase its equity
in the future; whether the deal included debt


financing; the interest rate charged; the activity
that was being financed; variables that indicate
whether the SBIC previously provided financing
to the firm; and whether the SBIC offered man­
agement services to the small business.
Figure 1 provides a comparison of several
measures of performance for our sample of
SBICs versus other financial institutions over
the 1986-91 period. In brief, SBICs performed
poorly over this period. Panel A of figure 1
shows that SBICs experienced very low ROEs
between 1986 and 1991 and performed worse
than commercial banks. SBICs’ returns on
equity were negative (-0.2 percent) over the
1986-91 period, and were positive for only
two of the six years. Panel B of figure 1 re­
ports the failure rates for sampled SBICs and
other financial institutions. The failure rate for
SBICs was a little above 11 percent per year,
compared with 5.5 percent for savings and loan
associations and 1 percent for commercial
banks." Over 56 percent of the 280 SBICs
were liquidated, had their licenses revoked, or
voluntarily surrendered their licenses prior to
the end of 1993.
Figure 2 shows that bank-owned SBICs
performed significantly better than their non­
bank-owned counterparts.12 Bank-owned
SBICs had a mean ROE of 1.9 percent over the
1986-91 period, while nonbank-owned SBICs
earned a -1.5 percent ROE. Failure rates dif­
fered as well: 41.4 percent of bank-owned
SBICs had failed by 1993, while the compa­
rable figure for nonbank-owned SBICs was
64.1 percent. The difference in failure rates is
even greater if failure is defined to include
only liquidations and license revocations.
Figures 3 and 4 show that SBA leverage
was used by a majority of the SBICs in our
sample, but it also reveals two other aspects of
SBA leverage usage. First, nonbank-owned
SBICs are much more likely to use SBA lever­
age than bank-owned SBICs (figure 3). Conse­
quently, the mean ratio of SBA funds to total
assets is much lower for bank-owned SBICs than
for nonbank-owned SBICs (figure 4, panel A).
Second, conditional on using any SBA lever­
age at all, bank-owned SBICs still used less
leverage than their nonbank-owned counter­
parts, and their usage declined over the period
under review (figure 4, panel B). It is clear from
these figures that, by and large, bank-owned
SBICs are not exploiting the SBA financing
subsidy to the same extent as other SBICs.




Performance of bank- and
nonbank-owned SBICs
A. Returns on equity, 1986-91

B. Failure rates, 1987-93
percent active at beginning of year that failed by year-end

Notes: Average ROEs = 1.9 percent for bank-owned
SBICs; -1 .5 percent for nonbank-ow ned SBICs; and
-0 .2 percent for all SBICs. Average failure rates = 41.4
percent for bank-owned SBICs; 64.1 percent for
nonbank-owned SBICs; and 56.1 percent for all SBICs.
Failure is defined as liquidation or revocation of an
SBIC's license by the SBA or surrender of license by
an SBIC.
Source: Authors' calculations from data provided by
the U.S. Sm all Business Adm inistration.

Factors affecting SBIC performance

Why should SBA leverage influence re­
turn on equity (ROE) and the likelihood of
failure, and what other factors may explain
SBICs’ weak earnings and failure? How might
access to SBA subsidies affect the returns on
capital invested in SBICs? One would expect
that borrowing money at a subsidized rate
would raise the returns to private investors. If
there are no market imperfections, then inves­
tors will invest in SBICs until their risk-adjust­
ed (post-subsidy) rates of return equal those
available in other financial intermediaries.
This means more projects would be funded


than would be the case in a world without SBA
subsidies. However, if only the riskiest SBICs —
those that would otherwise be unable to raise
funds or could do so only at a hefty risk premi­
um—use leverage, then this adverse selection
problem may mean we observe a positive rela­
tionship between failure and SBA leverage.
Further, if SBICs that use SBA leverage do
so because they intend to invest in riskier
projects than they would if only their own mon­
ey were at stake, this moral hazard may also
point to a positive relationship between failure
and leverage.


Finally, aside from these two informationrelated concerns, we consider the prepayment
effect and the mismatch effect. The SBA regu­
lations in effect during the period under review
essentially forbade prepayment of SBA-guaranteed debt during its first five years; hence,
SBA regulations matched the minimum dura­
tion of SBICs’ debt and the loans they made.
Thus, falling interest rates could mean a de­
cline in investment income but no commensurate
decline in interest expenses, putting pressure on
SBICs’ profits. This prepayment effect would
likely be most pronounced for SBICs with
large loan portfolios.13 A second factor is that
SBA leverage required regular interest pay­
ments to the SBA, whether or not the SBIC
earned any income over that period. Thus,
many SBICs, especially equity-oriented SBICs
whose realized income consists primarily of
variable capital gains, may have found SBA
leverage quite burdensome—the mismatch effect.
Overall, then, we have several reasons to expect
that SBA leverage may be negatively related to
ROE and positively related to failure.
The relationship between ROE (and failure)
and SBA leverage is obviously a complex one.
We consider three measures of SBA leverage.
The first measure, the ratio of total SBA funds
to total assets (SBATA), is a good indicator of
how an SBIC is funding its assets; that is,
whether it is funding a large or small fraction of
its assets with publicly subsidized funds. The
second leverage measure, the ratio of total SBA


funds to private capital (SBAPRIV), gives a sense
of the extent to which the SBIC’s own dollars
are at stake relative to subsidized dollars.
Thus, SBAPRIV may be a better measure of the
possibility of moral hazard problems arising.
The SBA implicitly recognized this possibility
when it developed regulations limiting the
amount of leverage to $3 of publicly subsidized
capital for every $1 of privately provided capi­
tal. Our third leverage measure, DSBATA, is
defined as the net change in SBA funding rela­
tive to total assets. Holding other things con­
stant, we expect that ROE should decrease and
the likelihood of an SBIC failure should in­
crease with SBA leverage. Thus,
where SBALEV captures the extent to which
an SBIC uses SBA funds; FAILURE is an
indicator variable which is equal to one if an
SBIC is liquidated, voluntarily surrenders its
license, or has its license revoked, zero other­
wise; CONTROL VARIABLES is a set of addi­
tional variables influencing ROE and SBIC
failure; and e and p are identically and inde­
pendently distributed error terms.
The bank failure literature suggests a set of
control variables that is likely to be important
in examining the relationship between SBA
leverage and performance, as measured by
profits or failure.14 We group these variables
as follows:
Asset composition and quality—The diver­
sification and quality of an SBIC’s asset port­
folio, as well as the share of loans in its securi­
ties portfolio, are likely to be related to profit­
ability (failure). PCOMP, the ratio of loans to
portfolio securities, is a crude measure that
controls for asset risk. SBICLOSS, the ratio of
loss provisions on accounts receivable to total
expenses, is a measure of asset quality and may
be negatively (positively) related to profitabili­
ty (failure). Two diversification measures,
HERFGEO and HERFSIC2, are Herfindahl
indexes constructed from the flows of invest­
ments made by the SBIC over the 1983-92
period; HERFGEO (HERFSIC2) is based on
flows by state (two-digit SIC industry) of the



small business receiving funding.15 High levels
of diversification (low Herfindahls) may be
associated with high profitability (low failure),
but specialization can yield economies on
monitoring costs incurred by the SBIC; conse­
quently, the net effect of the Herfindahls on
profitability and failure is uncertain. A related
measure is INSTATE, which is the share of
dollars invested by an SBIC in small businesses
located in its home state over the 1983-92
period. High levels of INSTATE may mean
lower monitoring costs, thus higher profits
(lower failure) for an SBIC.
Other SBIC characteristics—SBIC size
(SBICSIZE), as measured by the natural loga­
rithm of total assets (TA), and age (SBICAGE)
are control variables, though standard argu­
ments are that large SBICs may be more diver­
sified and may hire better managers than small
ones. We also include the ratio of operating
expenses to total assets (OPEX) to capture the
notion that efficient SBICs will earn superior
returns and be less likely to fail.
Characteristics of the small businesses
being financed—We consider two features, the
dollar-weighted mean age of the small business­
es receiving funding by the SBIC {AGEFIRM)\
and the share of dollar investments going to
firms with fewer than 50 employees (£1-49).
These measures also help to control for asset
risk, to the extent that smaller, younger firms are
riskier on average than are larger, older ones.
Projects being funded—We argue that the
types of projects funded by an SBIC are likely
to be correlated with its profitability (and fail­
ure). Each investment made by an SBIC is
identified as being intended to finance a certain
type of project being undertaken by the small
business receiving funding, for example, re­
search and development, land acquisition, or
operating capital. We grouped the ten possible
project types into three categories. USETRANS
is defined as the share of dollars invested in
transactions-type projects, whose execution is
likely to involve little managerial discretion by
the small business and to require little monitor­
ing by the SBIC. We include plant moderniza­
tion, debt consolidation, new building or plant,
machinery acquisition, and land acquisition
projects in this category. USERELAT is defined
as the share of dollars invested in relationship-type projects that are likely to involve
high levels of managerial discretion and SBIC


monitoring. We include acquisitions of existing
businesses, marketing, research and develop­
ment, and an other catch-all category here.
Finally, USEOPKAP is the share of dollars
invested in the last category, operating capital.

In principle, it is important to control for
the types of projects and financial characteris­
tics of the small businesses being financed by
SBICs when examining the relationship be­
tween SBA leverage and performance. Hence,


Characteristics of SBICs and their investments, 1986-91 means
Characteristics of SBICs


Total assets, $ m il.
Age, years
(Current assets-current liabilities)/total assets
(Provisions for losses on
accounts receivable)/total expenses
Net investment income/total assets
Operating expenses/total assets
(Loans/total securities), book value
(Total securities/total assets), market value
SBA funds/total assets
SBA funds/private capital
Growth rate of total assets, in logs
Cumulative realized profits net of
unrealized losses/private capital

Characteristics of SBICs



$-weighted mean age of firms
funded by an SBIC in each year
Share of invested funds in each year
intended for transactions-type projects
Share of invested funds in each year
intended for relationship-type projects
Share of invested funds in each year
intended for operating capital projects
Share of invested funds in each year
going to firm s with 1-49 employees
Share of invested funds in each year
going to firm s with 50-249 employees
Share of invested funds in each year
going to firm s with 250+ employees
Herfindahl index, based on ten-year flows by
tw o-digit SIC industry of small businesses
Herfindahl index, based on ten-year flows by
location (state) of small businesses
Share of invested funds going to small businesses
located in the same state as the SBIC
Share of invested funds in each year
going to firm s receiving funding for the
first time from this SBIC
Share of invested funds in each year
going to firm s in manufacturing sector
Share of invested funds in each year
going to firm s in transportation sector
Share of invested funds in each year
going to firm s in retail sector
Share of invested funds in each year
going to firm s in services sector


Banko w n ed

Nonbankow ned











Banko w n ed





0.135” '









0.563” '



0.348” '





























Nonbankow n ed

Notes: Sample is 280 SBICs, 1986-91. Total observations: 1,102. Means are unweighted. *, * * , and *** i ndicate means
for bank-owned SBIC differ significantly from means for nonbank-owned at the 10%, 5%, and 1% levels, respectively.
Source: Authors' calculations from data provided by the Small Business Administration.




in the empirical specifications of equations
1 and 2, we include many of these measures as
control variables.
Comparison of means—Table 1 reports
the mean values of selected variables for all
SBICs and for the bank- and nonbank-owned
SBICs over the 1986-91 period. First, com­
pared with nonbank-owned SBICs, bankowned SBICs were larger (SBICSIZE), more
equity-oriented (PCOMP), and more liquid
(SBICL1Q and ACOMP). Second, as described
above in more detail, bank-owned SBICs used
less SBA leverage (SBATA and SBAPRIV).
Third, they funded larger firms (£1-49 and
£50-249) and more relationship-oriented
projects (USERELAT). They also funded more
firms in the manufacturing and service sectors
and fewer in the transportation and retail sec­
tors. Finally, bank-owned SBICs grew much
more rapidly than did nonbank-owned SBICs
from 1986 to 1991 (.AGROW).
Performance of SBICs

The following equation provides a simple
empirical specification of the relationship
between ROE and selected financial variables:

R O E j.t = k 0n + y . t=2
T, k 0,t

DUM + £ 1 S B I C S I Z E


+ £ 2 S B I C A G Ej.t + £3 S B I C L O S S j.t
+ kA
P O R T F O L I O j.t + k 5, S B A L E Vj.t

+ £6 O P E X J.t + £1 A G E F I R M J.t
+ £ 8£ 7 —49J.t + £9 H E R F G E O j
+ £ io H E R F S I C 2 j + £ u, I N S T A T Ej + ej.r,

where j,t denotes SBIC j in year t, DUMt
(t = 2,3,...,£) are time-specific binary variables,
other explanatory variables are as defined
earlier (see table 1 and text); and e t is an error
term.16 PORTFOLIO is a vector of measures
of income-earning assets held by SBICs, and
we consider two alternative vectors detailed
below. We estimate equation 3 using timeseries cross-sectional data from 1986 to 1991
for the full sample of SBICs and for the bankand nonbank-owned subsamples of SBICs.
To determine the relationship between
failure of SBICs and our explanatory variables,
we estimate the following logit model by maxi­
mum-likelihood procedures:
4) Prob(FAILURE.; = 1) = <|>(X. f2 (3),
where FAILURE. r is equal to one if an SBIC is
liquidated, voluntarily surrenders its license, or



has its license revoked and zero otherwise; X „
is the vector of explanatory variables on the
right-hand side of equation 3; (3 is a vector of
parameter estimates for the independent vari­
ables X. f 2; and (j) is the log odds ratio.17
ROE results

Table 2 reports the results from regressing
ROE on our first SBALEV measure, SBATA,
and other variables, for the full sample as well
as separately for the bank-owned and nonbankowned SBICs. Column 1 contains the results
on the simplest model estimated over the full
sample of 280 SBICs, 1986-91, where the
PORTFOLIO vector includes USETRANS and
USERELAT. Two things stand out in column
1. First, the relationship between SBA lever­
age and ROE is negative, even after controlling
for SBIC age, size, and portfolio composition,
and characteristics of projects and small busi­
nesses. Second, several, though not all, of the
other variables are significantly related to ROE.
In particular, the operating expense variable,
OPEX, has a significant negative correlation
with ROE, and asset quality, as measured by
SBICLOSS, has a modest negative effect. The
share of investments going to transactions-type
projects and, to a lesser extent, the share going
to relationship-type projects are positively
correlated with ROE (recall that operating
capital is the excluded category). The diversi­
fication measures HERFGEO and HERFSIC2 are
not significant, nor are INSTATE, AGEFIRM, or
£1-49. Thus, there is little evidence that, once
portfolio characteristics are taken into account,
the types of small businesses funded by SBICs
are important correlates of profitability.
Columns 2 and 3, which report results
from the same regression estimated for the bank
and nonbank samples, show that SBICLOSS,
USETRANS, and USERELAT are important
only for the nonbank SBICs. Given that the
effect of the loss variable is likely to be nil for
SBICs whose portfolios contain mostly equities
(losses on accounts receivable are not likely to
be related to the ultimate quality of the equities
held by the SBIC) and that banks do most of
their investing in the form of equity, the
SBICLOSS result is not surprising. Why
USETRANS and USERELAT seem important
only for nonbank-owned SBICs is more of a
puzzle. An alternative specification is presented
in columns 4-6 of table 2; here, the USETRANS
and USERELAT variables are replaced by



The relationship between return on equity (ROE) and SBA leverage







-1 .0 3 ***








-0.24 2***





Notes: Sample is 280 SBICs, 1986-91. Dependent variable: ROE, 1986-91. Each specification includes (unreported) time
dummies, and standard errors are in parentheses below coefficient estimates. *, * * , and * * * indicate significance at the
10%, 5%, and 1% levels, respectively.
Source: Authors' calculations from data provided by the Small Business Administration.

PCOMP, the ratio of loans to securities at book
value. Since USETRANS and PCOMP are high­
ly correlated (SBICs tend to finance transactions-oriented projects with debt), we exclude
the USE variables in this specification. The
main result is unchanged: SBA leverage is
negatively related to ROE, even after control­
ling for other factors that may influence profit­
Next, we consider our two alternative
measures of SBA leverage, SBAPRIV and
DSBATA. The results from using SBAPRIV
shown in columns 1-3 of table 3 are quite
similar to the results using SBATA in table 2,
columns 4-6: SBA leverage has a significant
negative effect, though the statistical signifi­


cance of the effect is dampened with the new
measure. The regression results from using
DSBATA, in columns 4-6 of table 3, indicate
that increases in SBA leverage relative to total
assets affect ROE negatively only for nonbankowned SBICs, not bank-owned SBICs. When
considered in light of the SBA leverage usage
patterns described above, this result is not
surprising. Bank-owned SBICs were shedding
their already low levels of SBA leverage over
the 1986-91 period, while they were growing
rapidly and earning higher returns than nonbank-owned SBICs. The relationship between
leverage and ROE thus seems quite different
for the two types of SBICs.



The relationship between ROE and alternative measures of SBA leverage



-0.04 3***












-1 .0 2 ***





Notes: Sample is 280 SBICs, 1986-91. Dependent variable: ROE, 1986-91. Each specification includes (unreported) tim e
dummies, and standard errors are in parentheses below coefficient estimates. *, * * , and * * * indicate significance at the
10%, 5%, and 1% levels, respectively.
Source: Authors' calculations from data provided by the Small Business Administration.

Our principal finding from tables 2 and 3
is that after controlling for other factors that
can influence ROE, we still find a strong, neg­
ative relationship between SBA leverage and
profitability of SBICs. Can we identify which
of the stories sketched above is most impor­
tant? A report from the U.S. GAO (1993)
emphasizes both the mismatch effect and the
prepayment effect. To investigate the mis­
match story, we reestimated equation 3, adding
an interaction term to the set of regressors—the
product of SBATA and PCOMP. Our reason­
ing was that the sign of its coefficient would be
positive under the mismatch story, that is, the
negative effect of SBA leverage on ROE
would be most pronounced for SBICs with
low values of PCOMP (high shares of equi­
ties in their portfolios). In fact, we do obtain



a positive coefficient estimate on this interac­
tion term, offering some support for the mis­
match story.18
To investigate the prepayment effect, we
reestimated equation 3, allowing the coeffi­
cient on SBA leverage to vary over time. We
found statistically significant coefficients on
the time dummy-SBA leverage interaction
terms, suggesting that the prepayment story
may be important. Next, we considered three
possible ways of identifying the contribution
from prepayment restrictions, and we found
little evidence that prepayment restrictions
were the source of the negative leverage-ROE
relationship. Below, we briefly describe the
interest rate environment faced by SBICs dur­
ing our sample period and our findings on the
prepayment issue.


Interest rates were high in the early 1980s
compared with the years covered by our study,
1986-91. In the 1981-85 period, the ten-year
U.S. Treasury bond rate averaged 12.2 percent,
while over the 1986-91 period, it averaged
8.3 percent. If SBICs were unable to refinance
their existing high-rate debt in the early years
of our sample period, their profitability may
have been adversely affected. We argue that
this restriction, if important, should show up in
our analysis in any one of the following three
ways. First, the impact of SBA leverage on
ROE should vary depending on whether inter­
est rates are high or low relative to previous
years. When interest rates are falling, we
would expect the negative effect of SBA lever­
age to become more pronounced. To address
this, we reestimated the ROE equation of table
2, columns 4-6, adding an interaction term for
SBA leverage and the change in the ten-year
Treasury rate.19 We found a negative coeffi­
cient on the interaction term, so that when
interest rates were falling in the early years of
our sample, the negative impact of SBA lever­
age on ROE was mitigated, not exacerbated as
the prepayment story would imply.
A second prepayment story emphasizes
that the cost of failing to refinance high-rate
debt is that though liabilities remain expensive,
the assets of SBICs earn lower returns in the
lower interest rate environment. That is, if an
SBIC’s customers can refinance when rates fall
but the SBIC cannot, then the SBIC’s liabilities
remain costly, while its earnings on assets
decline. Under this story, a measure of the
interest rate spread earned by an SBIC would
be a narrower and better measure of the net
earnings likely to be affected by a decline in
interest rates. To investigate this, we reesti­
mated equation 3, now using an interest rate
spread as the dependent variable, including an
interaction term between SBA leverage and the
change in interest rates, and controlling for
macroeconomic conditions by including the
growth rate of real GDP.20 Again, we found no
evidence that leverage’s negative effect is most
pronounced when interest rates are falling.
Finally, we computed what each SBIC’s
interest expenses would have been had it refi­
nanced its entire stock of debt at the current
year’s ten-year Treasury rate. The prepayment
story implies that SBICs whose actual interest
expenses greatly exceeded these imputed ex­


penses (measured by the difference between
actual and computed interest expenses relative
to total assets) are those for whom the prepay­
ment restrictions are most burdensome; thus,
we should see low ROEs for these SBICs.
The simple correlation between ROE and this
difference measure is indeed negative.21 How­
ever in a regression of ROE on the same vari­
ables as in table 2 columns 4-6, plus this dif­
ference measure, the measure comes in strong­
ly significant but with a positive coefficient,
not a negative one. Again, this evidence does
not support the prepayment story.
In summary, we have little evidence that
the prepayment restrictions faced by SBICs
during our sample period are the main source
of the negative relationship between SBA
leverage and ROE. However, we do find some
support for the idea that the regular interest
payments due on SBA leverage adversely af­
fected profits at equity-oriented SBICs. More
research is needed to consider the relative
importance of other possible explanations for
the negative ROE-SBA leverage relationship.
Failure results

Table 4 reports the results from the estima­
tion of equation 4 for the full sample and the
bank- and nonbank-owned samples. The first
column for each sample presents the maximum
likelihood estimates of the parameters and their
standard errors. The second column reports the
marginal effects of the explanatory variables
on the probability of failure.
Consistent with the ROE results, SBA
leverage measured by SBATA is negatively
correlated with SBIC performance: SBICs
with higher SBATA have a higher probability
of failure two years hence. Furthermore, the
positive relationship between SBA leverage
and probability of failure is stronger for non­
banks. While an increase in SBATA increases
the probability of failure for a bank-owned
SBIC by 0.125, a similar increase in SBATA
increases the probability of failure for a non­
bank-owned SBIC by 0.187.
The correlations between failure and
SBICSIZE and SBICLOSS are also consistent
with the earlier results. SBICSIZE is negatively
correlated with the probability of failure in all
samples. SBICLOSS is positively correlated
with the probability of failure, but has a signif­
icant effect only for the full sample. In the full
and nonbank samples, higher ratios of loans to


T h e r e la tio n sh ip b e tw e e n th e p r o b a b ility o f fa ilu r e a n d S B A le v e r a g e



-0.37 5***





* 2<16>



















Notes: The dependent variable is an indicator variable that takes on a value of one if an SBIC failed two years hence;
otherwise, it takes on a value of zero. Failure is defined as either liquidation or revocation of license by the SBA, or
surrender of license by an SBIC. SBATA is the ratio of SBA funds divided by total assets. In addition to the above
explanatory variables, the model also includes tim e dumm ies for the years 1987-91. The MLE column presents the
m aximum likelihood estimates of the parameters and their standard errors. The PROB column presents the marginal
effects of the right-hand side variables (X) on the probability of failure, computed at the mean values of X. * , * * , and * * *
indicate statistical significance at 10%, 5%, and 1% levels, respectively.
Source: Authors' calculations from data provided by the Small Business Administration.

total portfolio securities (PCOMP) are associ­
ated with lower probabilities of failure. On the
other hand, PCOMP is not significant in the
bank-owned sample. This result is comparable
to the ROE results reported above.
Higher operating expenses are associated
with higher probabilities of failure, and this
relationship is particularly strong for the non­
bank-owned SBICs. Taken together with earli­
er results on ROE, these results indicate that
high operating expenses are associated with
low profitability contemporaneously for all
SBICs. For nonbank SBICs, high operating
expenses are also associated with poor long­
term performance, which suggests that the
consequences of operating inefficiencies at



nonbank-owned SBICs are more persistent.
Among the variables that describe the
investment strategy of SBICs, only the indus­
try-diversification measure, HERFSIC2, is
significantly related to probability of failure.
SBICs that are not diversified are more likely
to fail than well-diversified SBICs; however,
the relationship is significant only for the
bank-owned SBICs.
A l t e r n a t i v e v ie w s o f f a i l u r e

As Kane (1985, 1989) and others have
recognized, failure of institutions with access
to government liability-guarantees is not an
automatic consequence of a weakened financial
condition. It results from a conscious decision


by the regulatory agency to acknowledge and
act upon the weakened financial condition of
an institution. Our definition of SBIC failure
combines three different events, liquidation,
revocation, and surrender of license. Liquida­
tion and revocation are generally thought to be
choices of the SB A, while surrender of license
is a choice of the SBIC. How sensitive are our
results about SBA leverage to our definition of
failure? When we reestimated equation 4 on
the sample of SBICs consisting of survivors
and those who were liquidated during our sam­
ple period, we obtained results very similar to

the ones described above. However, using a
sample consisting of survivors and those who
surrender their licenses over the sample period
yields different results: SBA leverage is no
longer a statistically significant correlate of the
probability of failure, where failure is defined
as the surrender of a license.
The positive leverage-failure correlation
in the liquidation sample reflects both an eco­
nomic and a regulatory effect of leverage, and
without further work, we cannot disentangle
the two. Since leverage is not an important
correlate of failure in the surrenders-only sam-


The relationship between the probability of failure and SBA leverage,
including SBA’s measure of cumulative profitability










-1.13 4***











Notes: The dependent variable is an indicator variable that takes on a value of one if an SBIC failed tw o years hence;
otherwise, it takes on a value of zero. Failure is defined as either liquidation or revocation of license by the SBA, or
surrender of license by an SBIC. SBATA is the ratio of SBA funds divided by total assets. In addition to the above
explanatory variables, the model also includes tim e dummies for the years 1987-91. The MLE column presents the
m aximum likelihood estimates of the parameters and their standard errors. The PROB column presents the marginal
effects of the right-hand side variables (X) on the probability of failure, computed at the mean values of X. *, * * , and * * *
indicate statistical significance at 10%, 5%, and 1% levels, respectively.
Source: Authors' calculations from data provided by the Small Business Administration.



pie, a sample for which regulatory determi­
nants of failure were presumably not impor­
tant, the economic effect seems to be nil. How
can we reconcile this result with our claims
about the economic effects of leverage? First,
the distinction between liquidations and surren­
ders in practice is not as clear as our discussion
has implied. An SBIC may surrender its license
just before facing a certain liquidation action
by the SBA. Similarly, liquidations may occur
for purely economic reasons. For example, the
U.S. GAO (1993) reported that several SBICs
entered liquidation to avoid the prepayment
penalties associated with paying off their SBA
leverage. So, we do not view liquidations as
purely regulatory events, nor surrenders as
purely economic events. Second, we have
other evidence from our ROE analysis that the
negative effect of SBA leverage on perfor­
mance remains even when the sample consists
only of survivors and surrenders, that is, when
SBICs that ultimately are liquidated are removed
from the sample. Estimating equation 3 on this
other sample still yields a significant, negative
coefficient on SBA leverage, which is consis­
tent with there being an economic effect of
leverage on performance. In summary, though
we cannot gauge the quantitative importance of
the economic effects of leverage versus any
regulatory impact coming through the SBA’s
closure rule, we feel confident that the positive
coefficient on leverage in the failure equations
truly reflects the negative economic impact of
leverage on performance.
Finally, as noted earlier, the SBA consid­
ers an SBIC to be a poor performer if net real­
ized losses plus unrealized losses of the SBIC
exceed 50 percent of its private capital. If an
SBIC is capital impaired by this measure, the
SBA considers the SBIC in default and has the
right to liquidate its assets. Table 5 reports the
results from the estimation of equation 4 when
the SBA’s measure of performance, KIMPBA,
is included in the model as another explanatory
variable.22 The greater the SBA’s exposure to
losses, the more likely it is to take actions to
close an investment company. Thus, we expect
that the probability of SBIC failure will increase
with SBA leverage and with the degree of capi­
tal impairment.
We find that SBICs that perform well by
the SBA’s standards are indeed less likely to
fail; this relationship is particularly strong for



the nonbank-owned SBICs. For nonbankowned SBICs, including KIMPBA in the model
dampens the relationship between probability
of failure and SBA leverage. Because most of
the nonbank-owned SBICs take advantage of
SBA subsidies, it is not surprising that SBA
closure decisions are related more to the finan­
cial condition of these SBICs than to the level
of their SBA funding. On the other hand, SBA
leverage remains a significant correlate of
probability of failure for bank-owned SBICs,
even after KIMPBA is included. Since there
are significant differences across bank-owned
SBICs in the use of SBA funding, it is not
surprising that the level of SBA funding, as
well as their financial condition, is significant­
ly correlated with the probability of failure for
these SBICs.

Encouraging financial institutions to pro­
vide funding to small businesses has been a
central goal of U.S. public policy for a long
time. The SBIC program is designed to en­
courage the flow of long-term capital to small
firms. Because government guarantees are
used to fund many of the companies licensed
under the program, their performance is of
particular interest to policymakers.
In this article, we analyze the performance
of 280 SBICs that were active at the beginning
of 1986, paying special attention to the impact
of access to government liability guarantees on
ROE and failure. We find that SBICs performed
poorly. Of the 280 SBICs, over half had failed
by 1993. The ROE measure reveals a similarly
dismal performance.
We find that high usage of SBA-guaranteed
debt is associated with poor performance, partic­
ularly for nonbank-owned SBICs. We describe
several factors that may account for this rela­
tionship and offer evidence on two of them, the
prepayment effect and the mismatch effect. We
find little evidence that prepayment restrictions
faced by SBICs are important factors behind the
poor performance record of SBICs, but we do
find evidence that equity-oriented SBICs found
SBA leverage burdensome due to its regular
interest payment requirements. Our results are
also consistent with information-related prob­
lems (adverse selection and moral hazard)
being important. However, our results are not
sufficiently precise to differentiate these infor­
mation-related effects of leverage from its


other effects. Nevertheless, the results suggest
that public subsidies aimed at encouraging the
flow of funds to small firms may have unin­
tended consequences if the assets funded by
SBICs are riskier than they would have been in
the absence of the subsidy.
Finally, we note that in 1994 the SB A
revised many regulations pertaining to the
SBIC program. For example, minimum private
capital requirements were raised, prepayment
restrictions were lifted, and a new equity-like

form of leverage was developed and made
available to equity-oriented SBICs. Our analy­
sis suggests that the latter change may be quite
valuable and that lifting the prepayment restric­
tions may be less so. Furthermore, higher
capital requirements could, in principle, miti­
gate some of the information-related problems
that characterized the program in earlier years.
However, a complete assessment of the likely
impact of the new regulations on the perfor­
mance of SBICs must wait for future research.

'Initially, the Small Business Administration was estab­
lished as a temporary government agency to provide
intermediate-term financing to small firms. In 1958,
Congress made the SBA a permanent government agency.
For a discussion, see Osborn (1975).
T he SBA’s Statistical Package reports that 1,361 SBICs
were licensed over the 1959-94 period. Of these, 455 (33
percent) were transferred into liquidation between 1967
and 1994.
3For example, bank failures generated losses to the FDIC
of about $40 billion. For thrifts, the loss was near $200
billion, most of which was beyond the resources of the
deposit insurer and was thus charged to taxpayers. For a
discussion of the magnitude of the bank and thrift debacle
of the 1980s, see Bartholomew (1993) and Kaufman
(1995). Over the 1985-89 period, the cost to the FDIC to
close failing commercial banks averaged about 17 cents
per dollar of failed bank assets. See Barth, Brumbaugh,
and Litan (1992) for a discussion of resolution costs
associated with bank failures. For the now defunct Feder­
al Savings and Loan Insurance Corporation, the cost to
close failing S&Ls averaged about 33 cents per dollar of
assets over the 1985-89 period. See Barth (1991) for the
numbers used to compute the cost per dollar of assets.
4In 1994, the SBA put into effect new regulations that were
significantly different from those in effect over the 1986—
91 period. In this article, we focus on the regulation during
the 1986-91 period. In 1976, the program was extended to
include specialized SBICs (SSBICs) that provide funds to
small firms owned by “economically disadvantaged per­
sons.” In this article, we focus only on regular SBICs,
leaving an analysis of SSBICs for a future study.
5Under certain circumstances, SBICs can obtain up to $4
in SBA funds for every $1 of private capital, up to a
maximum amount of $35 million.
6The general partners are usually liable for all obligations
of a partnership. Thus, the liability structure offered by
the SBA is a departure from this norm and offers a relief
to general partners.
7If the SBIC provides a plan of divestiture, it can maintain a
controlling interest in a small business up to seven years.


"Limits on interest rates that can be charged to small
businesses are effective for all SBICs, whether or not they
use SBA leverage.
T he SBA’s SBIC Statistical Package reports that there
were 335 reporting SBICs in 1986.
l0Specifically, the financial statements pertain to the fiscal
years 1987-92.
"Our definition of SBIC failure is not exactly comparable
with that used for banks and savings and loan associations
(S&Ls). For SBICs, we define failure as liquidation,
revocation, or voluntary surrender of license. Few, if any,
banks or S&Ls voluntarily surrender their charters, and
the numbers in figure 1 exclude these voluntary surren­
ders. If our definition of SBIC failure included only
liquidations, the results would still indicate a higher
failure rate for SBICs.
,2An SBIC is classified as bank-owned in any year in
which at least 10 percent of its equity was controlled by a
banking organization. Otherwise, the SBIC is classified
as nonbank-owned.
l3This would be true if the mean duration of equity
investments was greater than the mean duration of debt
l4Sinkey (1975), Altman (1977), and Martin (1977)
analyze financial ratios constructed from balance sheets
and income statements to develop a system to help regula­
tors identify financially troubled institutions as early as
possible. These financial ratios were grouped into five
broad categories: capital adequacy, asset quality, manage­
ment competence, earnings, and liquidity. The same
types of broad categories were used by Avery and Hanweck (1984), Barth et al. (1985), Benston (1985), and
Gajewski (1989) to examine the likelihood of an institu­
tion’s closure. Cole (1993) examines economic insolven­
cy and closure using a larger number of financial factors
than in the previous studies. For an excellent review of
the literature on bank failure, see Demirgiic-Kunt (1989).
15The Herfindahl index is often used to measure competi­
tion in banking markets. It is calculated as the sum of the
squares of deposit shares of all competitors in a market.


If the index is equal to one, little or no diversification (or
competition) in the market is present, and the smaller the
index the more diversified (or competitive) the market.
Here, HERFS1C2, for example, is calculated as the sum of
squared shares of funding in a particular SIC code to the
total fundings made by an SBIC over the 1982-92 period.
Similarly, the shares of investments made by an SBIC by
state are used to calculate the HERFGEO index.
,6Recall that H' RFGEO and HERFSIC2 are computed
over the full ten-year period, 1983-92, as opposed to
separately for each year. Our method implicitly as­
sumes a ten-year duration for the investments made by
SBICs, whereas the year-by-year method assumes a
one-year duration.
l7Many failed SBICs are missing financial records for the
year preceding failure. Consequently, we focus on twoyear ahead failure prediction in the models we present
below. Once we discard the available observations per­
taining to the year before failure, as well as four observa­
tions with data problems, we have 1,102 observations, of
which 414 (688) are classified as bank-owned (nonbankowned) SBICs.
l8Our coefficient (standard error) estimates are -0.326
(0.044) on the SBATA variable and 0.119 (0.083) on the
SBATA-PCOMP interaction variable. At the sample
mean of PCOMP, which is 0.381, this implies a total

coefficient of -0.281 on SBATA\ for SBICs with zero
loans in their portfolios, the total coefficient is -0.326.
Analyzing bank-owned and nonbank-owned SBICs
separately, we find that the interaction coefficient is
positive and significant at the 1 percent level for only the
nonbank-owned SBICs.
'’We controlled for macroeconomic conditions by includ­
ing the growth rate of real GDP in this regression, as well
as in all the other regressions described in this section on
prepayment restrictions; thus, time dummies are not
included as in equation 3.
2l)We defined the interest rate spread as the difference
between the interest rate received by the SBIC (interest
income relative to interest-earning assets) and the interest
rate paid by the SBIC (interest expenses relative to total
debt owed by the SBIC).
2lWe recognize that we cannot exclude the possibility that
a large difference may occur for some SBICs because they
are currently poor performers that wish to avoid the
scrutiny associated with refinancing. Though the SBA
may not explicitly price risk when it sets interest rates on
its debentures, it may indirectly penalize a poorly per­
forming SBIC in other ways when the SBIC requests new
"This analysis uses our original definition of SBIC failure.

Altman, Edward I., “Predicting performance
in the savings and loan association industry,”
Journal of Monetary Economics, Vol. 3, No. 4,
October 1977, pp. 443-466.
Avery, Robert B., and Gerald Hanweck, “A
dynamic analysis of bank failures,” Proceed­
ings of a Conference on Bank Structure and
Competition, Federal Reserve Bank of Chica­
go, 1984, pp. 380-395.
Barth, James R., The Great Savings and Loan
Debacle, Washington, DC: The AEI Press,
Barth, James R., R. Dan Brumbaugh, Jr.,
and Robert E. Litan, The Future of American
Banking, New York: M. E. Sharpe, Inc., 1992.
Barth, James R., R. Dan Brumbaugh, Jr.,
Daniel Sauerhaft, and George H. K. Wang,
“Thrift institution failures: Causes and policy
issues,” Proceedings o f a Conference on Bank
Structure and Competition, Federal Reserve
Bank of Chicago, 1985, pp. 184-216.



Bartholomew, Philip F., Resolving the Thrift
Crisis, Washington, DC: Congressional Budget
Office, April 1993.
Benston, George J., “An analysis of the causes
of savings and loan association failure,” Mono­
graph Series in Finance and Economics, New
York University, Salomon Brothers Center for
the Study of Financial Institutions, 1985.
Brewer, Elijah III, and Hesna Genay, “Small
business investment companies: Financial
characteristics and investments,” Journal of
Small Business Management, Vol. 33, No. 3,
July 1995, pp. 38-56.
_________ , “Funding small businesses through
the SBIC program,” Economic Perspectives,
Federal Reserve Bank of Chicago, Vol. 18, No.
3, May/June 1994, pp. 22-34.
Cole, Rebel A., “When are thrift institutions
closed? An agency-theoretic model,” Journal
of Financial Services Research, Vol. 7, No. 4,
December 1993, pp. 283-307.


Demirgiic-Kunt, Asli, “Deposit institution
failures: A review of empirical literature,”
Economic Review, Federal Reserve Bank of
Cleveland, Vol. 25, No. 4, Quarter 4, 1989,
pp. 2-18.
Gajewski, George R., “Assessing the risk of
bank failure,” Proceedings of a Conference on
Bank Structure and Competition, Federal Re­
serve Bank of Chicago, 1989, pp. 432-456.
Kane, Edward J., The S&L Insurance Mess,
Washington, DC: Urban Institute Press, 1989.
__________, The Gathering Crisis in Federal
Deposit Insurance, Cambridge, MA: MIT
Press, 1985.
Kaufman, George G., “The U.S. banking
debacle of the 1980s: An overview and les­
sons,” The Financier: ACMT, May 1995,
pp. 9-26.

Osborn, Richard C., “Providing risk capital
for small business: Experience of the SBICs,”
Quarterly Review o f Economics and Business,
Vol. 15, No. 1, Spring 1975, pp. 77-90.
Sinkey, Joseph, “A multivariate statistical
analysis of the characteristics of problem
banks,” Journal o f Finance, Vol. 30, No. 1,
March 1975, pp. 21-36.
United States General Accounting Office,
“Better oversight of SBIC programs could
reduce federal losses,” Washington, DC: GAO,
report, No. T-RCED-95-285, September 1995.
__________, “Financial health of small busi­
ness investment companies,” Washington:
GAO, report, No. RCED-93-51, May 1993.
United States Small Business Administra­
tion, “SBIC statistical package,” 1995.

Martin, Daniel, “Early warning of bank fail­
ure: A logit regression approach,” Journal of
Banking and Finance, Vol. 1, No. 6, November
1977, pp. 249-276.



P u b lic In form ation C e n te r

Federal Reserve Bank of Chicago
P.O. Box 834
Chicago, Illinois 60690-0834


D o n o t fo rw a rd
A d d ress c o rre c tio n re q u es ted
R eturn p o stag e g u a ra n te e d

Mailing label corrections or deletions
Correct or mark Delete from mailing list on the
label and fax it to 312-322-2341, or send to:
Mail Services
Federal Reserve Bank of Chicago
P.O. Box 834
Chicago, Illinois 60690-0834